can boosting minority car-ownership rates narrow inter...
TRANSCRIPT
Can Boosting Minority Car-Ownership Rates Narrow Inter-Racial Employment Gaps?
Steven RaphaelGoldman School of Public PolicyUniversity of California, Berkeley
Michael StollSchool of Public Policy and Social Research
University of California, Los [email protected]
June 2000
This research is supported by a grant from the National Science Foundation, SBR-9709197, and aSmall Grant from the Joint Center for Poverty Research.
Abstract
In this paper, we assess whether boosting minority car-ownership rates would narrow inter-racialemployment rate differentials. We pursue two empirical strategies. First, we explore whether theeffect of auto ownership on the probability of being employed is greater for more segregated groupsof workers. Exploiting the fact that African-Americans are considerably more segregated fromwhites than are Latinos, we estimate car-employment effects for blacks, Latinos, and whites and testwhether these effects are largest for more segregated groups. Second, we use data at the level of themetropolitan area to test whether the car-employment effect for blacks relative to that for whitesincreases with the degree of black relative isolation from employment opportunities. We find thestrongest car effects for blacks, followed by Latinos, and then whites. Moreover, this ordering isstatistically significant. We also find that the relative car-employment effect for blacks is largest inmetropolitan areas where the relative isolation of blacks from employment opportunities is the mostsevere. Our empirical estimates indicate that raising minority car-ownership rates to the white carownership rate would eliminate 45 percent of the black-white employment rate differential and 17percent of the comparable Latinbo-white differential.
1For recent, thorough reviews of the spatial mismatch literature and, see Ihlanfeldt (1999)and Pugh (1998).
2Examples of such programs include the federal Empowerment Zones, the experimentalresidential mobility program �Moving to Opportunities� (MTO), and the Department ofTransportation�s �Access to Jobs� programs. For evaluations of the program effects of MTO, seeLudwig (1998) and Katz et. al. (2000). For a description of the Access to Jobs program andevaluation of the initial implementation, see GAO (1999). For an evaluation of the job creationeffects of state enterprise zone programs, see Papke (1993).
1. Introduction
Over the past three decades, considerable effort has been devoted to assessing the importance
of spatial mismatch in determining racial and ethnic differences in employment outcomes. The
hypothesis posits that persistent racial housing segregation in U.S. metropolitan areas coupled with
the spatial decentralization of employment have left black and, to a lesser extent, Latino workers
physically isolated from ever-important suburban employment centers.1 Given the difficulties of
reverse-commuting by public transit and the high proportions of blacks and Latinos that do not own
cars, this spatial disadvantage literally removes many suburban locations from the opportunity sets
of inner-city minority workers.
To the extent that mismatch is important, closing racial and ethnic gaps in employment and
earnings requires improving the access of spatially-isolated minority workers to the full set of
employment opportunities within regional economies. Improving accessibility can be accomplished
through some combination of community development, residential mobility, and transportation
programs.2 Among the latter set of options, a potential tool for enhancing accessibility would be to
increase auto access for racial and ethnic minorities. Racial differences in car-ownership rates are
large, comparable in magnitude to the black-white difference in home-ownership rates documented
by Oliver and Shapiro (1997). Moreover, car-ownership rates for low-skilled workers are quite
sensitive to small changes in operating costs (Raphael and Rice 2000), suggesting that moderate
2
subsidies may significantly increase auto access for racial and ethnic minorities.
In this paper, we assess whether boosting minority car-ownership rates would narrow inter-
racial employment rate differentials. We pursue two empirical strategies. First, we explore whether
the effect of auto ownership on the probability of being employed is greater for more spatially
isolated groups of workers. The literature on racial housing segregation clearly demonstrates that
blacks are highly segregated from the majority white population (Massey and Denton, 1993) and in
a manner that spatially isolates blacks from new employment opportunities (Stoll et. al. 2000).
Latino households are also segregated, though to a degree considerably less than the level of
segregation between blacks and whites (Massey and Denton 1999). If mismatch reduces minority
employment probabilities, and if auto-ownership can partially undo this effect, the employment
effect of auto ownership should be greatest for the most segregated workers. We test this proposition
Using microdata from the Survey of Income and Program Participation (SIPP).
Second, we assess whether the differences in the car-employment effect between black and
white workers increases with the severity of spatial mismatch. If spatial mismatch yields a car-
employment effect for black workers that is larger than that for white workers, then the black-white
difference in the car-employment effect should be larger in metropolitan areas where blacks (relative
to whites) are particularly isolated from employment opportunities. We test this proposition using
data from several sources. From the 1990 5 % Public Use Micro Data Sample (PUMS),we estimate
the black-white difference in the car-employment effect for 242 metropolitan areas in the U.S. Next,
we construct corresponding metropolitan-area measures of the relative spatial isolation of black
workers from employment opportunities using data from the 1992 Economic Census and zip-code
population counts from the 1990 Census of Population and Housing. We then test for a positive
3
3Stoll (1999) analyzing a sample of adults in Los Angeles and Holzer et. al. (1994)analyzing a national sample of youths show that car owners search greater geographic areas andultimately travel greater distances to work than do searchers using public transit or alternativemeans of transportation.
relationship between these two metropolitan-area level variables.
We find strong evidence that having access to a car is particularly important for black and
Latino workers. We find a difference in employment rates between car-owners and non car-owners
that is considerably larger among black workers than among white workers. Moreover, the car-
employment effect for Latino workers is significantly greater than the comparable effect for non-
Latino white workers yet significantly smaller than the effect for black workers. Finally, the
difference between the car-employment effect for black workers and white workers is greatest in
metropolitan areas where the relative isolation of black workers is most severe. Our estimates
indicate that raising minority car ownership rates to the car ownership rate for whites would narrow
the black-white employment rate differential by 45 percent and the comparable Latino-white
differential by 17 percent.
2. Urban Mismatch and Auto Access
The proposition that having access to a reliable car provides real advantages in terms of
finding and maintaining a job is not controversial. In most U.S. metropolitan areas, one can
commute greater distances in shorter time periods and, holding distance constant, reach a fuller set
of potential work locations using a privately-owned car rather than public transit.3 For low-skilled
workers, being confined to public transit may seriously worsen employment prospects for a variety
4
4Hamermesh (1996) analyzes the likelihood of working irregular hours in the U.S. Botheducation and age have strong negative effects on the probability of working shifts from 7PM to10PM and 10PM to 6AM for both men and women. Hence, the young and the less educated aremore likely to work non-traditional schedules. Black men are also significantly more likely towork these irregular hours, while for women there is no effect of race.
5Holzer et. al. (1994) find that youths with cars experience shorter unemployment spellsand earn higher wages than youths without cars. Ong (1996) analyzes a sample of welfarerecipient residing in California and finds substantial differences in employment rates and hoursworked between those with cars and those without. O�Regan and Quigley (1999) find large car-employment effects for recipients of public aid using data from the 1990 decennial census.
of reasons. Such workers are more likely to work irregular hours4 while public transit schedules tend
to offer more frequent service during traditional morning and afternoon peak commute periods. This
incongruity in schedules may result in longer commutes, a relatively high probability of being late,
or both.
Moreover, the residential location choices of low-skilled workers are likely to be
geographically constrained by zoning restrictions limiting the location and quantity of low-income
housing. Such constraints may limit the ability of low-skilled workers to choose residential locations
within reasonable public-transit commutes of important employment centers. In light of these
considerations, it is not surprising that researchers have found large differences in employment rates
between car-owners and non car-owners.5
For minority workers, residential location choices are particularly constrained by relatively
low incomes and pervasive racial discrimination in housing rental and sales markets (Yinger 1995).
Moreover, the existing mismatch literature clearly demonstrates that low- and semi-skilled
employment opportunities are scarce in minority neighborhoods relative to the residential
concentration of low- and semi-skilled labor (Stoll et. al. 2000). In addition, several authors have
demonstrated intra-metropolitan patterns of employment growth that favor non-minority
5
6In fact, Holzer et. al. (1994) find larger effects of car-access on unemployment spells forblack youth relative to white youth.
E A S Bi i i i i= + + +α α α ε1 2 3 . (1)
∆
∆ ∆ ∆
B
B BA
BS
E E B C E E B CE A B C E A B CE S B C E S B C
= = = − = == = = − = = +
= = − = =
= −
( | , ) ( | , )[ ( | , ) ( | , )][ ( | , ) ( | , )]
1 1 1 01 1 1 01 1 1 0
1
2
1 2
αα
α α
(2)
neighborhoods (Mouw forthcoming, Raphael 1998, Stoll and Raphael 2000). Hence, one might
argue that having access to a car would be particularly important in determining the employment
outcomes of minority workers.6
These ideas can be formalized with a simple linear probability model of employment
determination. Assume that the categorical variable, Ei, indicating whether individual i is employed
depends on individual skills, Si, and one�s spatial accessibility to employment locations, Ai. Spatial
accessibility is akin to the density of one�s employment opportunity set, where accessible
employment opportunities are defined as those jobs within a reasonable commute distance from
one�s residential location. We assume that both accessibility and skills positively affect the
probability of being employed according to the linear equation
where gi is a mean-zero, randomly distributed disturbance term and Bi is a dummy variable indicating
a black worker.
Car ownership (denoted by the indicator variable, Ci) affects the probability of being
employed by improving accessibility � i.e., car owners can access a greater proportion of a
metropolitan area�s labor market than can non-car owners. In terms of the variables in the model,
this assumption implies that E(A |B, C=1) > E(A|B, C=0). For black workers, the expected
difference in employment rates between car owners and non-car owners is given by the expression
6
7A strategy for addressing omitted-variables bias as well as the possibility of reversecausality would be to find exogenous determinants of car-ownership and use these variables asinstruments in a 2SLS model of employment determination. Raphael and Rice (2000) pursuethis strategy using inter-state variation in gas taxes and average car-insurance premiums asinstruments for car ownership. They find car-employment effects that are large, statisticallysignificant, and comparable in magnitude across OLS and 2SLS models. Hence, after adjustingfor variables readily available in most microdata sets, there is little evidence of omitted-variablesor simultaneity bias in simple OLS estimates of car-employment effects.
where ∆AB is substituted for the expected accessibility difference between black car owners and non
car-owners and ∆SB is substituted for the comparable expected skill differential. The �true� car effect
for black workers is given by the first term (the improvement in accessibility multiplied by the
marginal effect of accessibility) while the second term provides that portion of the mean difference
in employment rates between black car owners and non-car owners due to inherent productivity
differences.
As is evident from equation (2), assessing the real effect of car access on the probability of
being employed requires statistically distinguishing the portion of the employment rate differential
caused by improved accessibility from the portion of the differential reflecting differences in average
skill endowments between those with and without cars. One approach to tackling this issue would
estimate an adjusted employment difference between car owners and non-car owners holding
constant all relevant factors that determine employment and differ systematically across these two
groups of workers. Unfortunately, the set of covariates included in most micro-data sources is likely
to be incomplete and, hence, such regression-adjusted estimates of the car-employment effect may
be biased by the omission of important unobservable factors.7
Fortunately, a lower-bound estimate of the car-employment effect for blacks that addresses
omitted-variables bias can be computed by comparing the employment rate differential in equation
7
∆ ∆ ∆ ∆ ∆ ∆B W BA
WA
BS
WS− = − + −α α1 2( ) ( ), (3)
∆ ∆ ∆ ∆B W BA
WA− = −α 1( ). (4)
(2) to a comparable differential for white workers. Define ∆w as the employment rate difference
between car owners and non-car owners for white workers comparable to the difference for black
workers defined above. Subtracting this difference for white workers from that for blacks yields the
expression
where ∆Aw and ∆S
w are the expected differences in accessibility and skill endowments between white
workers with and without cars. Assuming that the skill differential between car owners and non-car
owners is comparable across races (i.e., ,∆SB = ∆S
w) the double-difference in equation (3) reduces to
This final expression gives the differential effect of cars on the probability of being employed caused
by racial differences in the accessibility boost of having access to a car.
Equation (4) is a lower-bound estimate of the car-employment effect for black workers since
it differences-away the accessibility improvement realized by white car owners. If we were to
assume that the entire employment rate differential between white car owners and white non-car
owners was due to unobservable heterogeneity (that is to say, ∆Aw = 0, ∆S
w >0), then equation (4)
provides an accurate estimate of the black car-employment effect. This, however, is unlikely. For
reasons discussed above, even the residents of jobs-rich suburban communities are likely to benefit
from access to a car. Morever, instrumenting for car-ownership in linear employment probability
models estimated on representative samples of the U.S. working-age population yields positive
significant estimates of the car-employment effect that are comparable to simple regression-adjusted
8
car effect estimates (Raphael and Rice 2000). This suggests that on average, cars exert positive
causal effects on the probability of being employed. Nonetheless, using lower bound estimates of
the car-employment effect for blacks should partially mitigate concerns about omitted variables bias.
The quantity in equation (4) will be greater than zero if two conditions are satisfied. First,
accessibility must matter (i.e., α1 >0). Otherwise, there would be no employment benefit to car-
ownership. Second, the accessibility benefits of owning a car must be greater for blacks than for
whites -- i.e, ∆AB > ∆A
w. This latter condition may fail to hold for several reasons. First, blacks may
be no more spatially isolated from employment opportunities than are whites, and hence, there would
be no differential benefit associated with having access to a car -- i.e., spatial mismatch is not an
important contributor to black-white inequality. Alternatively, the spatial isolation of blacks may
be so extreme that even having access to a car does not in any way neutralize the deleterious
employment consequences of mismatch. If this were the case, there may still be some benefit to car-
access for both black and white workers, but there would be no differential improvement in
accessibility for black workers. Hence, testing for a positive double-difference estimate as described
by equation (4) provides a rather strict test of the mismatch hypothesis.
The simple double-difference framework outlined in equations (1) through (4) form the basis
for the empirical tests that we implement below. We now turn to making these arguments
operational, outlining specific hypotheses, and assessing the relative contributions of mismatch and
differences in car ownership rates to the inter-racial employment rate differential.
3. Empirical Strategy and Data Description
The arguments presented in the previous section posit that the effect of auto access on the
9
8In all models, we define exclusive racial/ethnic categories � i.e., non-Latino black, non-Latino white, and Latino.
probability of being employed should be larger for more spatially isolated workers. Here, we outline
two specific empirical strategies designed to assess this proposition. Our first strategy exploits the
differences in the extent of segregation between backs and whites and between Latinos and whites.
Both blacks and Latinos are residentially segregated from the majority non-Latino white population.
In addition, the intra-metropolitan patterns of segregation are similar, with both Latinos and blacks
more likely to reside in older inner-city and inner-ring suburban communities. However,
conventional segregation indices show that blacks are much more segregated, and in turn, spatially
isolated from high-growth suburban employment centers, than are Latinos. Hence, if car-ownership
partially neutralizes the adverse employment effects of being spatial isolated, we would expect the
largest employment differentials between those with and without cars for black workers, the next
largest differential for Latinos, and the smallest differential for non-Latino white workers.
We estimate the double-difference car effect in equation (4) using a black-white comparison,
a black-Latino comparison, and a Latino-white comparison.8 The simplest test of the mismatch
hypothesis would be the test of whether the black-white double-difference estimate is positive and
statistically significant. The more stringent test of the mismatch hypothesis would be to test for
positive significant double-difference estimates in the black-white and Latino-white comparisons,
as well as a positive significant effect in the black-Latino comparison. Affirmative findings in all
three comparisons would suggest that the ordering of the car-employment effects is statistically
significant.
To be sure, the key assumption identifying equation (4) (that the skill differentials between
10
9In fact, tabulations from the 1992 and 1993 SIPP indicate that the average difference ineducational attainment and age between those with and without cars is slightly larger for whitesthan for blacks (by one-tenth of a year for educational attainment and approximately half a yearfor age).
E B C C B Xi i i i i i i= + + + + +β β β β δ ν0 1 2 3 * , (5)
car-owners and non-car owners are equal across race and ethnicity) is strong. While we feel that
there are no reasons a priori to suspect that these skill differentials vary across racial and ethnic
groups,9 if the assumption is violated the double-difference estimate in equation (4) may not be fully
purged of the effects of skills. For example, if the skill differentials between car owners and non-car
owners are larger for blacks than for whites, the estimate of the differential car effect would be
biased upward, since not all of the difference in skills is differenced-away. Alternatively, the skill
differential between car owners and non-car owners may be larger for whites than for blacks. In this
scenario, the double-difference in equation (4) would �over-adjust� for skill differentials and
underestimate the differential boost that blacks receive from car ownership above and beyond the
effect on white employment rates.
One way to partially address this concern would be to estimate a regression-adjusted double-
difference estimate that holds constant those human capital and demographic characteristics that are
observable. For the black-white comparison, an adjusted double-difference comparable to that in
equation (4) comes from estimating the equation
where all observable determinants are included in the vector Xi, and the adjusted double-difference
is given by the coefficient β3 on the interaction term between the indicator variables for car owners
and black workers. This coefficient measures the extent to which the car-employment effect for
11
10Since the data are drawn from the fourth waves of each panel, the figures correspond tothe years 1993 for the 1992 panel and 1994 for the 1993 panel.
blacks exceeds that for whites. Holding constant all observable variables, the identification
assumption reduces to assuming comparable differentials across racial and ethnic groups in
unobserved skills between those with and without cars. Below, we present estimates of both the
unadjusted double-difference in equation (4) and the adjusted double-difference estimate in equation
(5).
We estimate equations (4) and (5) using microdata from the fourth waves of the 1992 and
1993 Survey of Income and Program Participation (SIPP). These surveys provide large nationally
representative samples that include standard labor force participation, demographic, and human
capital variables. In addition, the fourth wave topical modules of the SIPP collect information on
up to three cars per household, including the age of the automobile, the financing status, and the
person identifier of the car owner within the household. We use this latter variable to explicitly
identify individuals that own a car rather than individuals residing in a household where someone
owns a car. We restrict the sample to civilians, 16 to 65 years of age, with no work-preventing
disabilities. In addition, we further restrict the sample to individuals that are either white, black, or
Latino. Given that the survey collects complete information on all household automobiles only for
those households with 3 or fewer cars, we restrict the sample throughout to individuals residing in
such households. After taking into account the other sample restrictions, this restriction eliminates
approximately 6 percent of the observations.
Table 1 presents car ownership rates for whites, blacks, and Latinos calculated from the
combined 1992 and1993 SIPP samples.10 The table presents figures for the three racial/ethnic groups
12
Table 1Car-Ownership Rates by Race/Ethnicity, Educational Attainment, and Age 1993/1994
White Black Latino
All 0.757 (0.002) 0.468 (0.007) 0.522 (0.007)
Educational Attainment Less than 12 years 12 years 13 to 15 years 16 years More than 16 years
0.507 (0.007)0.767 (0.004)0.773 (0.004)0.823 (0.005)0.873 (0.005)
0.284 (0.014)0.455 (0.011)0.526 (0.015)0.700 (0.023)0.740 (0.027)
0.435 (0.012)0.520 (0.013)0.611 (0.018)0.714 (0.029)0.746 (0.036)
Age 16-19 20-24 25-34 35-44 45-54 55-65
0.143 (0.006)0.522 (0.008)0.803 (0.004)0.870 (0.004)0.891 (0.004)0.874 (0.005)
0.036 (0.007)0.206 (0.017)0.489 (0.014)0.612 (0.014)0.679 (0.018)0.705 (0.021)
0.088 (0.012)0.330 (0.019)0.589 (0.014)0.693 (0.014)0.685 (0.020)0.638 (0.026)
Standard errors are in parentheses. The sample combines the fourth wave of the 1992 and 1994Survey of Income and Program Participation.
overall and for the three groups stratified by educational attainment and age. As is evident, there are
large and statistically significant inter-racial and inter-ethnic differences in car ownership rates. For
all whites in our sample, 76 percent own cars, compared with 47 percent of blacks, and 52 percent
of Latinos. Moreover, within educational attainment categories whites have higher (and statistically
distinguishable) car ownership rates than do blacks and Latinos. For example, 51 percent of whites
with less than 12 years of education own cars, compared with 28 percent of blacks and 44 percent
of Latinos with comparable educations. Similarly, among individuals with 16 plus years of
schooling, 87 percent of whites, 71 percent of blacks, and 64 percent of Latinos own cars.
The largest racial/ethnic differences in car ownership rates occur for the relatively young
workers in our sample. For example, the black/white difference in car ownership rates are
13
11For this strategy we focus on the black-white comparisons only due to the fact that inmany PMSAs, the numbers of Latino observations are prohibitively small.
approximately 11 percent for those 16 to 19 years of age, 31 percent for those 20 to 24, and 31
percent for those 25 to 34. The Latino/white differences for these age groups are also large, though
smaller than the differences between blacks and whites. Hence, to the extent that owning a car has
real employment effects, the large differences evident in Table 1 indicate that closing these gaps may
narrow inter-racial employment differentials.
Our first empirical strategy infers differential spatial isolation by assuming that segregation
from whites and being spatially-isolated from employment opportunities are synonymous. Based
on this indirect inference, we then test for an interaction between the car-employment effect and
mismatch by comparing the car effects for groups that differ with respect to their degree of
residential segregation. An alternative approach would directly measure the degree of spatial
isolation from employment and test for a positive relationship between empirically observed car
effects and the direct measure of mismatch. Our second empirical strategy takes this form.
Specifically, for the black-white comparisons only,11 we estimate the adjusted double-
difference car effect (equation 5) separately for 242 U.S. Primary Metropolitan Statistical Areas
(PMSAs) using data from the 5% Public Use Microdata Sample (PUMS) of the 1990 Census of
Population and Housing. We restrict the PUMS sample to civilian black and white observations that
are 16 to 65 years of age with no work-preventing disabilities. Unlike the SIPP, the census only
identifies whether someone in the household owns a car. Hence, our estimates of the car effects
using the PUMS are based on this less precise household level measure of auto-access. This PMSA-
level measure of the double-difference car effect is now our dependent variable.
14
12Define Blacki as the black population residing in zip-code i, Employmenti as the numberof jobs located in zip-code i, Black as the total black population in the metropolitan area, andEmployment as the total number of jobs in the metropolitan area. The dissimilarity scorebetween blacks and jobs is calculated using the equation, D = 3|Blacki/Black -Employmenti/Employment|, where the summation is over all zip-codes in the metropolitan area.
13We set net new jobs to zero in zip-codes experiencing net employment losses. Thistends to overstate the economic health of predominantly black zip-codes, since blacks are morelikely to reside in zip-codes with net job loss than are whites.
Next, we construct race-specific, PMSA-level measures of spatial isolation from employment
opportunities. Using zip-code level, place-of-work employment data from the 1992 Economic
Census and zip-code population counts from the 1990 Census Summary Tape Files 3B, we construct
MSA-level indices by race that measure the imbalance between residential distributions and
employment distributions. Specifically, we estimate jobs/people dissimilarity indices for four
employment measures.12 The dissimilarity index ranges from zero to one and can be interpreted as
the proportion of people (or jobs) that would have to move to yield a perfectly even distribution of
persons and jobs across zip codes within the metropolitan area. For example, our dissimilarity index
between blacks and retail jobs in Chicago is 0.74, while the comparable dissimilarity index for
whites is 0.28. These figures indicate that 74 percent of blacks and 28 percent of whites would have
to move (across zip codes) to be spatially distributed in perfect proportion with the spatial
distribution of retail employment.
We construct jobs/people dissimilarity indices for blacks and whites separately using four
separate zip-code level measures of employment: the 1992 levels of retail employment, the 1992
levels of service employment, new retail jobs added between 1987 and 1992, and new service jobs
added between 1987 and 1992.13 For each employment measure, we subtract the white/jobs
dissimilarity index from the black/jobs dissimilarity index to arrive at a PMSA-level measure of the
15
14We cannot calculate dissimilarity indices for the full 272 PMSAs identified in the 5percent PUMS due to differences in geography between the Economic Census and Census ofPopulation and Housing. The thirty metropolitan areas that we are missing are generally smallerareas with relatively small black populations. The figures presented in Table 4 are weighted bythe black populations of the MSAs. Hence, these figures indicate the isolation experienced bythe typical black resident in these 242 PMSAs relative to whites.
15White dissimilarity indices are larger than black indices for 30 retail level comparisons,22 retail growth comparisons, 51 service level comparisons, and 37 service growth comparisons. All PMSAs where black indices exceed comparable values for whites are small metropolitanareas with small black populations.
isolation of blacks from employment opportunities relative to the spatial isolation of whites. This
is our key explanatory variable. If mismatch is important, and if having a car partially undoes the
consequences of mismatch, then the relative employment effect of car-access for blacks should be
largest in those metropolitan areas where blacks are most isolated (relative to whites) from
employment opportunities.
Table 2 presents weighted averages of our jobs/people dissimilarity indices for 242 PMSAs.14
All four measure indicate that blacks are more segregated from employment opportunities than are
whites. Moreover, the black-white differences in the dissimilarity indices are highly statistically
significant in all cases. Comparisons of individual cities indicates that, for the most part, the
jobs/people dissimilarity indices are uniformly higher for blacks than they are for whites. Table A1
presents such comparisons for the twenty metropolitan areas with the largest black populations in
1990 (accounting for roughly 60 percent of the black metropolitan population in this year). In all
comparisons, black dissimilarity indices exceed white dissimilarity indices. For all 242 PMSAs, the
overwhelming majority of comparisons indicate that black dissimilarity indices exceed white
dissimilarity indices.15 Hence, these figures suggest strongly that African-American in the United
States have near uniformly inferior access to jobs relative to whites. What remains to be seen is
16
Table 2Mean Dissimilarity Scores Measuring Segregation Between Population and EmploymentOpportunities for Metropolitan Areas Identified in the 1990 PUMS
Blacks/Jobs Indices Whites/Jobs Indices Difference (Black-White)
Retail Dissimilarity Indices
Levels, 1992 Net Growth, 1987 to 1992
0.59 (0.007)0.81 (0.006)
0.31 (0.003)0.63 (0.006)
0.28 (0.008)0.18 (0.005)
Service DissimilarityIndices
Levels, 1992 Net Growth, 1987 to 1992
0.62 (0.008)0.75 (0.007)
0.42 (0.004)0.57 (0.005)
0.21 (0.008)0.18 (0.006)
Standard errors are in parentheses. Each figure is the mean for the 242 PMSAs for which we wereable to calculate double-difference car effects. The figures are weighted by the number of blackobservations observed in each PMSA. The levels dissimilarity index is calculated using zip-code levelinformation on the number of jobs located in the zip-code in 1992 and the number of people of therelevant race residing in the zip-code in 1990. The net growth indices uses net job growth between1987 and 1992, setting growth to zero for zip codes that lose employment over this time period. Information on population vby zip code comes from the 1990 Census of Population and HousingSummary Tape Files 3b. Information on job counts by zip codes comes from the Economic Census for1987 and 1992.
whether the benefits of auto access for blacks is largest in metropolitan areas where relative isolation
is the greatest.
4. Empirical Results
A. Inter-Group Comparisons of the Car-Employment Effect
Table 3 presents employment rate tabulations using data from the two SIPP surveys. The
table provides employment rates by race and ethnicity for all individuals in each sub-group,
employment rates for those with and without cars, and the difference in employment rates between
car owners and non-car owners. Starting with employment rates in the first row by race and
17
Table 3Employment Rates by Race/Ethnicity and Car-Ownership States and the Unadjusted Double-Difference Estimates
White Black Latino ∆2Black-White ∆2
Black-Latino ∆2Latino-White
All 0.746 (0.002) 0.631 (0.007) 0.619 (0.007) - - -
With CarWithout Car
0.800 (0.002)0.580 (0.005)
0.833 (0.008)0.453 (0.010)
0.765 (0.009)0.460 (0.011)
--
--
--
Difference 0.220 (0.005) 0.380 (0.013) 0.305 (0.014) 0.160 (0.013) 0.075 (0.019) 0.085 (0.013)
Standard errors are in parentheses. The data come from combining the fourth waves of the 1992 and 1994 Survey of Income andProgram Participation.
18
ethnicity, blacks and Latinos have considerably lower employment rates than whites. The overall
white employment rate exceeds the black employment rate by approximately 11 percentage points
and exceeds the Latino employment rate by roughly 13 percent. These differences, however, are
either non-existent or much smaller among workers with cars. Blacks with cars actually have a
higher employment rate (0.833) than whites with cars (0.800) while Latinos with cars have a slightly
lower employment rate (0.765) than both blacks and whites. For those who do not own cars, the
racial and ethnic employment rate differentials are quite pronounced. Specifically, among workers
without cars, the white employment rate exceeds the blacks employment rate by nearly 13 percentage
points and the Latino employment rate by 12 percentage points.
These patterns translate into larger car-employment effects for blacks and Latinos than for
whites. The bottom row of the table (corresponding to the first three columns) presents unadjusted,
group-specific estimates of the car-employment effect. The difference in employment rates between
those with and without cars is 22 percentage points for whites, 38 percentage points for blacks, and
31 percentage points for Latinos. Recall, the spatial mismatch hypothesis predicts that the effect of
car access should be largest for those workers who are most isolated from employment opportunities.
To the extent that segregation from whites proxies for spatial isolation, the patterns evident in Table
3 confirm this prediction.
To test whether these differences in the car-employment effect are statistically significant,
the last three columns of Table 3 present calculations of three unadjusted double-difference estimates
(corresponding to equation (4) of the previous section) and the accompanying standard errors. The
first double-difference subtracts the white car effect from the black car effect, the second subtracts
the Latino car effect from the black car effect, while the final estimate subtracts the white car effect
19
16For each year of the SIPP, we created 45 state dummy variables. We cannot createdummy variables for the full 50 states due to the fact that the SIPP aggregates some states withsmall populations into larger groups.
from the Latino car effect. All three double-difference estimates are positive and highly significant.
Hence, the larger car-employment effect for blacks is statistically distinguishable from the Latino
and white employment effects and the larger Latino car effect is statistically distinguishable from the
white car-employment effect.
To be sure, the figures and double-difference estimates presented in Table 3 do not adjust for
differences in skills and other characteristics that affect labor market outcomes and that may differ
inter-racially and between those with and without cars. To account for this possibility, Table 4
presents regression-adjusted estimates of the differential car effects comparable to the specification
in Equation (5). For the three comparisons (black/white, black/Latino, and Latino/white), we
estimate two linear-probability employment models. The first specification controls only for
race/ethnicity, having access to a car, and the interaction between these two dummy variables. The
coefficient on the interaction term is the unadjusted double-difference estimate and corresponds
directly to the calculations presented in Table 3. The next specification adds controls for several
variables available in the SIPP including age, gender, educational attainment, marital status, whether
the individual is enrolled in school, and whether there is an infant present in the household. The
model also includes 90 dummy variables for state groups, hence adjusting for any differences in the
state economy that might affect employment probabilities.16
The first two regressions present results for the black/white comparisons, the next two
regressions present results for the black/Latino models, while the final two regressions present results
for the Latino/white comparisons. The results indicate that controlling for observable characteristics
20
Table 4Linear Probability Employment Models that Provide the Regression-Adjusted Double-DifferenceEstimates of the Effects of Car Ownership on Minority Employment Prospects
Black-White Comparisons Black-Latino Comparisons Latino-White Comparisons
(1) (2) (3) (4) (5) (6)
Blacka
(Latino)-0.127(0.004)
-0.147(0.009)
-0.007(0.013)
-0.049(0.014)
-0.120(0.009)
-0.088(0.010)
Car 0.219(0.005)
0.134(0.005)
0.305(0.013)
0.188(0.014)
0.219(0.005)
0.136(0.006)
Black*Cara
(Latino*Car)0.160
(0.013)0.179
(0.013)0.075
(0.019)0.077
(0.018)0.085
(0.013)0.090
(0.012)
Female - -0.115(0.004)
- -0.141(0.009)
- -0.133(0.004)
Married - -0.061(0.005)
- -0.027(0.011)
- -0.059(0.004)
Years ofSchooling
- 0.018(0.001)
- 0.023(0.002)
- 0.018(0.001)
Age - 0.042(0.001)
- 0.039(0.002)
- 0.041(0.001)
Age2 - -0.0005(0.0000)
- -0.0005(0.0000)
- -0.0005(0.0000)
Infant - -0.106(0.007)
- -0.093(0.013)
- -0.109(0.007)
In School - -0.170(0.007)
- -0.194(0.015)
- -0.174(0.007)
R2 0.072 0.196 0.128 0.251 0.063 0.191
N 40,847 40,847 9,301 9,301 40,580 40,580
Standard errors are in parentheses. The models including the control variables in columns (2), (4), and(6) also include 90 state dummy variables accounting for the 45 state groupings provided in the SIPP forthe two separate years of the sample.a. For the Latino-White comparisons, the label in parentheses applies.
21
does not affect the size of the inter-racial/ethnic estimates of the relative car effects. In the
black/white comparisons, adjusting for observable characteristics yields a slight increase in the
relative car effect from 0.160 to 0.179. Similarly, in the black/Latino and Latino/white models,
adjusting the models leads to slight increases in the relative car effects. Notably, the base car effect
declines considerably (for example, from 0.219 to 0.134 in the black/white comparisons), indicating
that a significant portion of the overall employment rate differential between car owners and non-car
owners is explained by differences in observable human capital variables. The fact that the relative
car effects (the coefficients on the interaction terms) are not affected, however, suggests that the
relatively large effects for blacks and Latinos are not driven by differential selection biases by race
and ethnicity.
The results in Tables 3 and 4 combined with the figures on car-ownership rates in Table 1
can be used to estimate the proportion of the black-white and Latino-white employment rate
differentials that would be eliminated by raising minority car-ownership rates up to that for whites.
We start by making the conservative assumption that the entire base car effect in regressions (2) and
(6) captures unobserved skill differentials between car owners and non-car owners (and by extension,
that there is no employment effect of car ownership for whites). Under this assumption, the
differential effects for blacks and Latinos present lower bound estimates of the impact of car
ownership on the probability of being employed for members of these groups. Hence, multiplying
the difference in car ownership rates between blacks and whites by the differential effect of car
ownership provides a lower bound estimate of the effect on black employment rates of eliminating
the racial gap in car ownership rates.
The figures in Table 3 indicate a black/white employment rate differential of 11.5 percentage
22
points and a Latino/white differential of 12.7 percentage points. Assuming that having access to a
car increases the probability of being employment for blacks by 0.179 (estimate from regression (2)),
eliminating the approximate 30 percentage point black/white difference in car ownership rates would
narrow the black/white employment rate gap by 5.2 percentage points. This equals nearly 45 percent
of the black/white employment rate differential. A similar calculation indicates that eliminating the
Latino/white difference in car-ownership rates would close the Latino/white employment rate
differential by 2.1 percentage points. This amounts to 17 percent of the Latino/white differential
observed for the sample.
The results in Tables 3 and 4 indicate that the low car-ownership rates of blacks and Latinos
explain a fair portion of the black/white and Latino/white employment rate differentials. Moreover,
the results presented here provide support for the hypothesis that auto access matters considerably
more for workers that are more physically isolated from employment opportunities. In these tests,
we draw indirect inferences about the interactions between mismatch and the effect of auto access
based on the assumption that being residentially segregated from whites is equivalent to being
spatially isolated from employment opportunities. We now turn to a direct test of the relationship
between relative car effects and spatial mismatch.
B. Cross-Metropolitan Area Evidence
Here we present the results from our cross-metropolitan area analysis using data from the
1990 5% PUMS and the 1992 Economic Census. Recall, the analysis here focuses exclusively on
the black/white relative car effect due to the small number of Latino observations in many
metropolitan areas. We begin by describing the model we use to estimate the PMSA-specific
relative effects. Appendix Table A2 provides regression results for two linear employment
23
17Two minor differences in the PUMS specifications include the fact that we do notcontrol for the presence of an infant in the household and that we add an indicator variableindicating a work-limiting disability.
probability models using the entire 5% PUMS. We restrict our sample to black and white survey
respondents 16 to 65 years of age with no work preventing-disabilities. The models estimated are
quite similar to the model specifications estimated with the SIPP data in Table 4, with a few notable
exceptions. First, it is impossible to identify the exact owners of automobiles in the Census. Hence,
our proxy for auto access is coded to one if someone in the household owns a car. Second, our
specification of the effect of education is slightly different due to the categorical nature of the
educational attainment variable in the PUMS.17 Again, we first present estimation results with a
black indicator, a car access indicator, and the interactions between the two. We then add the other
covariates to the specification.
The results in Appendix Table A1 using the full national sample correspond closely with the
results using the SIPP sample. Access to a car has a much larger effect for blacks than for whites.
Moreover, adjusting for observable covariates does not appreciably alter the size of the relative car
effect. One difference from the SIPP results, however, concerns magnitude. While in the SIPP data
we find a differential car employment effect for blacks of approximately 18 percentage points, we
find a 10 percentage point differential effect using the PUMS sample. We attribute this disparity to
the difference in the way auto access is measured with the two samples. Surely, identifying the
actual owner of the car is a more precise measure of auto access than the household counterpart.
We estimate the second specification separately for each of 242 PMSAs. The PMSA-specific
coefficients on the interaction term between the black indicator variable and car access dummy from
these regressions provides our dependent variable. Our principal tests entail bivariate regressions
24
18We also ran regressions of the double-difference car effect on the ratio of the black-to-white jobs/people dissimilarity indices. This specification yields nearly identical results to thosein the figures. In addition, we estimated models where the black and white dissimilarity indiceswere entered separately rather than as differences. Using these alternative models, we tested theimplicit parameter restriction of the difference model that the coefficient on the white/jobsdissimilarity index is equal to the negative of the coefficient on the black/jobs dissimilarityindex. In all four models we failed to reject this restriction, thus, supporting the specificationdepicted in Figures 1 and 2.
19We also estimated the models in Figures 1 and 2 not weighting the data. This uniformlyleads to larger and more statistically significant coefficient estimates.
of the PMSA-level relative car effects on the PMSA-level black-white differences in the four
jobs/people segregation indices described above. Figures 1 and 2 present the results from these
bivariate regressions. Figures 1A and 1B present scatter plots of the double-difference car effects
against the black-white differences in the retail employment level indices and the retail employment
growth indices, respectively. Figures 2A and 2B provide similar scatter plots using the black-white
differences in the service level and service growth dissimilarity indices.
In each scatter plot we include the regression line as well as the coefficient estimates and R2
from a weighted regression of the double-difference car effects on the differences in dissimilarity
indices.18 We weight each regression by the number of black observations for the PMSA used to
compute the double-difference estimate.19 The relative weight placed on each observation is
indicated by the size of the bubble in the scatter plot.
Before discussing the regression results, we should highlight a few notable aspects of the
distributions of the explanatory and dependent variables that are revealed in the scatter plots. First,
in all four figures, the mass of the distribution of observations lies to the right of the vertical axis,
thus indicating that in nearly all metropolitan area (with the exception of a hand full) the blacks/jobs
25
-0 .2
-0 .1
0
0 .1
0 .2
0 .3
0 .4
-0 .2 -0 .1 0 0 .1 0 .2 0 .3 0 .4 0 .5 0 .6
B la ck-W h ite D iffe re n ce in R e ta i l Em p lo y m e n t D issim ila ri ty In d e x
Dou
ble-
Diff
eren
ce C
ar E
ffect
A . U s in g th e R e ta il E m p lo y m e n t D is s im ila rity In d ic e s
-0 .2
-0 .1
0
0 .1
0 .2
0 .3
0 .4
-0 .2 -0 .1 0 0 .1 0 .2 0 .3 0 .4 0 .5 0 .6
B l a c k -W h ite D i ffe re n c e i n R e ta i l E m p lo y m e n t G ro w th D issim i l a ri ty In d e x
Dou
ble-
Diff
eren
ce C
ar E
ffect
B . U s in g th e R e ta il E m p lo y m e n t G ro w th D is s im ila rity In d ic e s
Figure 1: Scatter Plots of the Double-Difference Car Effects Against Black-White Differencesin the Retail Dissimilarity Indices
Notes: The size of the bubbles are proportional to the number of black observations used to compute the double-difference car effects. The depicted regression lines are weighted by the number of black observations per PMSA.
26
-0.2
-0.1
0
0.1
0.2
0.3
0.4
-0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6
Bla ck-W hite D iffe re n ce in the S e rvice Em ploym e nt D issim ila ri ty Ind e x
Dou
ble-
Diff
eren
ce C
ar E
ffect
A . Usin g the S e rvice Em p loym e nt D issim ila rity Ind ice s
-0.2
-0.1
0
0.1
0.2
0.3
0.4
-0.4 -0.2 0 0.2 0.4 0.6
Bla ck-W hite Diffe re nce in Service Em ploym ent Grow th Dissim ila rity Inde x
Dou
ble-
Diff
eren
ce C
ar E
ffect
B. Using the Service Em ploym ent Grow th Dissim ila rity Indice s
Figure 2: Scatter Plots of the Double-Difference Car Effects Against Black-White Differencesin the Service Dissimilarity Indices
Notes: The size of the bubbles are proportional to the number of black observations used to compute the double-difference car effects. The depicted regression lines are weighted by the number of black observations per PMSA.
27
dissimilarity indices exceed the whites/jobs dissimilarity indices. Moreover, for those metropolitan
areas where the reverse is true (leading to negative black-white differences in the dissimilarity
indices) black populations are quite small (as is evident from the small bubbles used to mark the
observations). Hence, the basic plots in Figures 1 and 2 demonstrate, in an alternative manner, the
nearly uniform inferior access of blacks to employment opportunities relative to the accessibility
enjoyed by whites.
Concerning the distribution of the dependent variable, the mass of observations lies above
the horizontal axis. This indicates that in all but a few metropolitan areas, the effect of car
ownership on the employment rates of blacks exceeds the comparable effects for whites. Moreover,
the size of the bubble plots where the reverse is true (white car effects are larger than black car
effects yielding adjusted double-differences that lie below the horizontal axis) is generally small.
This has two implications. First, in those cities where we observe larger car effects for whites, there
are few black observations, thus yielding imprecise estimates of the double-difference car effect.
Hence, the smaller car effects for blacks in these instances may be attributable entirely to
measurement error. Second, given the small number of PMSAs lying below the horizontal axis and
the small numbers of black survey respondents accounted for by these observations, the figures
indicate that the overwhelming majority of blacks residing in metropolitan areas live in regions
where owning a car matters more for blacks than for whites (as measured by our adjusted double-
differences).
In Figures 1A and 1B, there are clear positive relationships between the PMSA-level relative
car effects and the relative isolation of blacks from retail employment opportunities. The coefficient
on the difference in dissimilarity indices is positive and significant for both the retail levels indices
28
20Recall, our dependent variable measures the differential car effect for blacks aftereliminating the base car effect for whites.
(p-value of 0.019) and the retail growth indices (p-value of 0.005). In Figures 2A and 2b, we also
observe a statistically significant positive relationship between the relative car effects and differences
in the service employment dissimilarity indices, though the relationships are a bit weaker than those
for the retail indices (the difference in the service level indices is significant at the7 percent level of
confidence while the difference in the service growth indices is significant at the 6 percent level of
confidence). Hence, these simple bivariate relationship indicate a clear positive relationship between
the relatively large car-employment effects for blacks and the degree of relative spatial isolation from
employment opportunities.
One might argue that the bivariate regressions presented in Figures 1 and 2 do not control
for possible selection across metropolitan areas along personal and human capital characteristics that
may be driving these significant relationships. However, the double-differences used as the
dependent variable are already purged of the effect of educational attainment, age, and the other
covariates listed in Appendix Table 2A, as well as any inter-PMSA sorting that is occurring among
white workers.20 Moreover, since the regressions used to generate the dependent variable where
estimated separately for each metropolitan area, the relative car effect estimates have also been
purged of any possible differences in the returns to observable covariates (in terms of the marginal
effects on employment probabilities) that may vary across PMSAs. For these reasons, we feel that
the bivariate regressions presented in the figures are not being driven by systematic variation across
PMSAs in relative skill differentials or difference in some other relevant personal characteristics.
Nonetheless, there still may be omitted metropolitan area characteristics that coincide with
29
21We also experimented with controlling for average population density and differences inmean travel times between blacks and whites. Including these variables did not alter the effect ofthe spatial isolation variable on the relative car effects. We also estimated models interacting theracial difference in dissimilarity indices with the proportion commuting by private auto and thetotal land area. The interaction terms were not significant in any of the regressions.
racial differences in spatial isolation from employment. For example, the quality of public transit
may vary from area to area or the total area covered by the PMSA may vary. While we do not have
extensive controls and PMSA level characteristics, we do have a few measures that we add to the
specifications of the models in Figures 1 and 2. Table 5 presents weighted regression results where
the dependent variable is the PMSA-level adjusted double-difference. For each dissimilarity index
we estimate two specifications: the first controlling for the racial difference in dissimilarity scores
only, and the second adding the proportion of PMSA workers that commute to work by private auto
(calculated from our 5% PUMS sample), and the total land area of the PMSA.21 The first eight
models present separate regressions for the four dissimilarity indices while the final two models
control for all of the dissimilarity scores in the same specification.
With the exception of the two regressions for the black-white difference in retail growth
dissimilarity scores, adding these two variables to the specification increases the point estimates of
the effect of relative black spatial isolation on the relative employment effect of owning a car. In the
models that add the two additional variables to the specification, the relative isolation measures exert
positive statistically significant (at the one percent level) effects in all four specifications. Hence,
the bivariate results survive (and in three of the four cases, are strengthened by) adding additional
covariates to the models. Controlling for all four dissimilarity scores at the same time yields rather
imprecise point estimates. This is not too surprising considering that the four measures are highly
30
Table 5Regression of the Adjusted Double-Difference Car Effect on the Black-White Differences in the Dissimilarity Indices Measuring SegregationBetween Population and Employment Opportunities
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Retail Dissimilarity Indices Black-White Difference in 1992 Levels
0.055(0.023)
0.081(0.024)
- - - - - - 0.061(0.056)
0.129(0.058)
Black-White Difference in 1987 to 1992 Net Growth
- - 0.132(0.037)
0.101(0.040)
- - - - 0.121(0.045)
0.011(0.055)
Service Dissimilarity Indices Black-White Difference in 1992 Levels
- - - - 0.047(0.026)
0.068(0.026)
- - -0.036(0.069)
-0.093(0.069)
Black-White Difference in 1987 to 1992 Net Growth
- - - - - - 0.061(0.032)
0.084(0.032)
-0.015(0.052)
0.049(0.054)
Proportion Commuting to Workby Private Auto
- 0.080(0.021)
- 0.043(0.022)
- 0.072(0.021)
- 0.070(0.021)
- 0.079(0.025)
Land Areaa - 0.011(0.012)
- 0.010(0.012)
- 0.011(0.012)
- 0.013(0.012)
- 0.012(0.012)
R2 0.022 0.091 0.049 0.071 0.014 0.072 0.015 0.073 0.056 0.100
F-statisticb
(P-value)- - - - - - - - 3.481
(0.008)3.502
(0.008)
N 242 242 242 242 242 242 242 242 242 242
Standard errors are in parentheses. All regression include a constant and are weighted by the number of black observations used to calculate the double-difference.a. Land area is measures in tens of thousands of acres.b. This row presents the test-statistics and p-values from a test of the joint significance of the four segregation indices.
31
correlated with one another. Nonetheless, F-test of the cumulative significance of all four measures
fails to reject the hypothesis that all of the coefficient are zero.
Concerning the performance of the two additional variables, the relative effect of car
ownership for blacks is higher in metropolitan areas that are more car dependent. The effect of this
variable is significant at one percent in all specifications with the exception of fourth regression
(significant at the five percent level of confidence). Finally, land area does not exert a statistically
measurable effect in any of the models.
In summary, the results in Figures 1 and 2 and Table 5 strongly confirm the proposition that
the relative importance of auto access on the employment prospects of blacks is more important in
metropolitan areas where blacks are more spatially isolated from employment opportunities than are
whites. Moreover, the positive effect of relative isolation on the relative car employment effect
survives additional controls for metropolitan area characteristics.
5. Conclusion
The results of this paper clearly indicate that having access to a car has disproportionately
large effects on the employment rates of workers that are spatially isolated from employment
opportunities. We find the largest car-employment effects for blacks, the next largest for Latinos,
and the smallest effects for whites. Moreover, we find strong evidence that the disparity between
the blacks and white car-employment effects is greatest in metropolitan areas where the relative
isolation of blacks from employment opportunities is most severe. Given the large differences in
car-ownership rates that we document, these results indicate that lack of access to transportation
plays a large role in explaining black/white, and to lesser degree, Latino/white differences in
32
employment rates. By extension, these results also suggest that subsidizing car-ownership may be
an effective policy tool for narrowing these employment gaps.
To be sure, employment policies that increase auto ownership rates will also increase the
externalities associated with increased private-auto work commutes and non-work trips. Nearly all
metropolitan areas in the United States suffer from traffic congestion, given the practically
insurmountable challenge of optimally pricing road usage. Increasing auto-ownership through, for
example, a subsidy to operating costs will surely increase traffic congestion. In addition, more autos
will certainly translate into more air pollution.
There are reasons, however, to suspect that increasing auto access for blacks and Latinos
would not add appreciably to congestion and pollution. Concerning the congestion externality, since
a disproportionately large share of blacks and Latinos live in central regions of metropolitan areas,
those individuals who commute to jobs located within city centers are unlikely to increase congestion
on inbound freeway routes. Moreover, those who locate employment in the suburbs will have
commutes that are in the reverse direction of the largest peak-period flows. O�Regan and Quigley
(1999) have made a similar point quite decisively in their discussion of the possible congestion
consequences of increasing car-ownership rates among welfare recipients. Another factor limiting
the addition to congestion costs concerns the fact that many of these individuals work non-standard
schedules and, hence, would be making private-auto commuted at times of the day when the external
costs of an additional trip are low. Finally, even an extreme policy that raises minority car ownership
rates to the level of whites would purchase new autos for a minority of a minority of the U.S.
working age population. Hence, both the congestion and pollution externalities caused by such
policies are likely to be small.
33
An important issue that we have not addressed in this study concerns the reasons why car-
ownership rates are so low for blacks and Latinos. Possible explanations include high insurance and
parking costs in inner-city communities, differential access to capital markets, low incomes, and
price discrimination in the markets for new and used cars. Sorting out the relative importance of
these alternative factors, and any other relevant variables for that matter, would greatly aid in
designing cost-effective policies that increase auto access.
Finally, the results presented here do not provide enough information to compare the relative
efficacy (in terms of alleviating inner-city employment problems) of community development
initiatives, residential mobility programs, training programs, and policies designed to increase
automobile accessibility. Of course, to the extent that all such policies alleviate the spatial imbalance
between labor supply and demand, these policy tools may be thought of as complements rather than
substitutes, with the effects of one initiative increasing the probability of success of alternatives.
Nonetheless, a careful comparative analysis of the marginal benefits per dollar spent may indicate
that certain policy options dominate. The strong results presented here indicate that transportation
policies geared towards fostering greater auto access should most definitely be considered in any
comparative benefit-cost analysis of policy initiatives designed to alleviate the spatial concentration
of joblessness.
34
References
Government Accounting Office (1999), Welfare Reform: Implementing DOT�s Access to JobsProgram in Its First Year, Report # GAO/RCED-00-14, Government Accounting Office.
Hamermesh, Daniel S. (1996), Workdays, Work Hours, and Work Schedules: Evidence for theUnited States and Germany, W.E. Upjohn Institute for Employment Research: Kalamazoo,Michigan.
Holzer, Harry J.; Ihlanfeldt, Keith R.; and David L. Sjoquist (1994), "Work, Search, and TravelAmong White and Black Youth," Journal of Urban Economics, 35: 320-345.
Ihlanfeldt, Keith R. and David L. Sjoquist (1998), �The Spatial Mismatch Hypothesis: A Review ofRecent Studies and Their Implications for Welfare Reform,� Housing Policy Debate, 9(4): 849-892.
Katz, Lawrence; Liebman, Jeffrey; and Jeffrey Kling (2000), �High Poverty Public Housing,Housing Vouchers, and Health: Evidence from a Randomized Mobility Experiment,� unpublishedmanuscript.
Ludwig, Jens (1998) �The Effects of Concentrated Poverty on Labor Market Outcomes: Evidencefrom a Randomized Experiment,� unpublished manuscript.
Massey, Douglas S. and Nancy A. Denton (1989), �Hypersegregation in U.S. Metropolitan Areas:Black and Hispanic Segregation Along Five Dimensions,� Demography, 26(3): 373-391.
Massey, Douglas S. and Nancy A. Denton (1993), American Apartheid: Segregation and the Makingof the Underclass, Harvard University Press: Cambridge.
Mouw, Ted (2000), �Job Relocation and the Racial Gap in Unemployment in Detroit and Chicago1980-1990: A Fixed-Effects Estimate of the Spatial Mismatch Hypothesis,� forthcoming AmericanJournal of Sociology.
Oliver, Melvin L. and Thomas M. Shapiro (1997), Black Wealth/White Wealth: A New Perspectiveon Racial Inequality, Routledge: New York and London.
Ong, Paul (1996), "Work and Automobile Ownership Among Welfare Recipients," Social WorkResearch, 20(4): 255-262.
O�Regan, Katherine M. and John M. Quigley (1999), �Spatial Isolation of Welfare Recipients: Whatdo we Know?,� Program on Housing and Urban Policy Working Paper #W99-003.
Pugh, Margaret (1998), �Barriers to Work: The Spatial Divide Between Jobs and Welfare Recipientsin Metropolitan Areas,� The Brookings Institution Center on Urban and Metropolitan Policy
35
Discussion Paper.
Papke, Leslie E. (1993), �What Do We Know About Enterprise Zones?,� NBER Working Paper#4251.
Raphael, Steven (1998), �The Spatial Mismatch Hypothesis of Black Youth Joblessness: Evidencefrom the San Francisco Bay Area,� Journal of Urban Economics, 43(1): 79-111.
Raphael, Steven and Lorien Rice (2000), �Car Ownership, Employment, and Earnings,� unpublishedmanuscript.
Stoll, Michael A. (2000), �Spatial Job Search, Spatial Mismatch, and the Employment and Wagesof Racial and Ethnic Groups in Los Angeles,� Journal of Urban Economics, 46(1): 129-155.
Stoll, Michael A.; Holzer, Harry J.; and Keith R. Ihlanfeldt (2000), �Within Cities and Suburbs:Racial Residential Concentration and the Spatial Distribution of Employment Opportunities AcrossSub-Metropolitan Areas,� Journal of Policy Analysis and Management, 19(2): 207-232.
Stoll, Michael A. and Steven Raphael (2000), �Racial Differences in Spatial Job Search Patterns:Exploring the Causes and Consequences,� forthcoming, Economic Geography.
Yinger, John (1995), Closed Doors, Opportunities Lost: The Continuing Costs of HousingDiscrimination, Russell Sage Foundation: New York.
36
Appendix Table A1Dissimilarity Scores Measuring Segregation Between Population and Employment forthe 20 Metropolitan Areas with the Largest Black Population in 1990
Retail Dissimilarity Indices Service Dissimilarity Indices
Levels, 1992Net Growth,1987 to 1992 Levels, 1992
Net Growth,1987 to 1992
Black White Black White Black White Black White
AtlantaBaltimoreBirminghamCharlotteChicago
0.590.570.600.470.74
0.330.290.420.350.28
0.800.880.730.850.89
0.530.670.560.750.67
0.650.600.640.480.79
0.460.400.540.460.42
0.770.840.710.780.86
0.490.630.620.620.56
ClevelandDallasDetroitHoustonLos Angeles
0.670.530.790.570.66
0.260.310.260.310.28
0.840.850.940.760.88
0.570.680.630.490.76
0.710.630.800.640.70
0.370.460.420.470.44
0.810.680.920.720.84
0.440.530.690.560.68
MemphisMiamiNew OrleansNew YorkNewark
0.540.600.490.710.69
0.330.250.350.360.30
0.800.820.690.870.89
0.550.580.600.770.76
0.600.670.550.770.70
0.420.340.410.540.35
0.820.850.660.820.83
0.590.560.500.610.60
NorfolkOaklandPhiladelphiaSt. LouisWashington, D.C.
0.430.610.720.670.56
0.310.280.290.290.35
0.690.790.910.830.82
0.550.680.680.590.64
0.460.610.750.720.64
0.360.350.420.410.48
0.470.740.830.800.75
0.410.500.600.540.62
The levels dissimilarity index is calculated using zip-code level information on the number of jobslocated in the zip-code in 1992 and the number of people of the relevant race residing in the zip-codein 1990. The net growth indices uses net job growth between 1987 and 1992, setting growth to zerofor zip codes that lose employment over this time period. Information on population by zip codecomes from the 1990 Census of Population and Housing Summary Tape Files 3b. Information on jobcounts by zip codes comes from the Economic Census for 1987 and 1992. Approximately sixtypercent of the 1990 black population living in metropolitan areas resided in one of the twenty PMSAslisted above.
37
Appendix Table A2Linear Probability Employment Models Using the 1990 5% PUMS and the Household Level Car-Ownership Variable
(1) (2)
Black -0.159(0.002)
-0.128(0.002)
Car 0.132(0.001)
0.104(0.001)
Black*Car 0.115(0.002)
0.103(0.001)
Female - -0.149(0.0003)
Married - 0.016(0.0001)
High School Graduate - 0.113(0.0005)
Some College - 0.164(0.0005)
College Graduate - 0.180(0.0007)
College + - 0.213(0.0008)
Age - 0.045(0.0001)
Age2 - -0.0006(0.0000)
In School - -0.104(0.0006)
Disabled - -0.047(0.0009)
R2 0.018 0.162
N 4,455,814 4,455,814
Standard errors in parentheses.